Article addresses that somewhat

> On the one hand, this output is not very surprising. AI researchers who create AI language models like the kind that power ChatGPT have known for years that these models can synthesize realistic permutations of information learned from those texts. It's how every AI assistant today works.

> But what makes this episode especially interesting is that a small hobbyist model trained by one man appears to have surprised him by reconstructing a coherent historical moment from scattered references across thousands of documents, connecting a specific year to actual events and figures without being explicitly taught these relationships. Grigorian hadn't intentionally trained the model on 1834 protest documentation; the AI assembled these connections from the ambient patterns in 6.25GB of Victorian-era writing.

This still doesn't seem wildly surprising. The event happened in 1834, he asked it a question about 1834 - why are we surprised, exactly, that the text-completion engine completed text in the expected manner?